151.4.2 Characterizing Uncertainty
‘Uncertainty’ is a complex and multifaceted property, sometimes originating in a lack of information, and at other times from quite fundamental disagreements about what is known or even knowable (Moss and Schneider, 2000). Furthermore, scientists often disagree about the best or most appropriate way to characterize these uncertainties: some can be quantified easily while others cannot. Moreover, appropriate characterization is dependent on the intended use of the information and the particular needs of that user community. Scientific uncertainty can be partitioned in various ways, in which the details of the partitioning usually depend on the context. For instance, the process and classifications used for evaluating observational uncertainty in climate science is not the same as that employed to evaluate projections of future change. Uncertainty in measured quantities can arise from a range of sources, such as statistical variation, variability, inherent randomness, inhomogeneity, approximation, subjective judgement, and linguistic imprecision (Morgan et al., 1990Morgan, M. G., M. Henrion, and M. Small, 1990: Uncertainty: A Guide to Dealing with Uncertainty in Quantitative Risk and Policy Analysis. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 332 pp. ), or from calibration methodologies, instrumental bias or instrumental limitations (JCGM, 2008JCGM, 2008: JCGM 100: 2008. GUM 1995 with minor corrections. Evaluation of measurement data—Guide to the expression of uncertainty in measurement. Joint Committee for Guides in Metrology.). In the modelling studies that underpin projections of future climate change, it is common to partition uncertainty into four main categories: scenario uncertainty, due to uncertainty of future emissions of GHGs and other forcing agents; ‘model uncertainty’ associated with climate models; internal variability and initial condition uncertainty; and forcing and boundary condition uncertainty for the assessment of historical and paleoclimate simulations (e.g., Collins and Allen, 2002Collins, M., and M. R. Allen, 2002: Assessing the relative roles of initial and boundary conditions in interannual to decadal climate predictability. J. Clim., 15, 3104–3109.; Yip et al., 2011Yip, S., C. A. T. Ferro, D. B. Stephenson, and E. Hawkins, 2011: [ A simple, coherent framework for partitioning uncertainty in climate predictions]. J. Climate, 24, 4634–4643.). Model uncertainty is an important contributor to uncertainty in climate predictions and projections. It includes, but is not restricted to, the uncertainties introduced by errors in the model’s representation of dynamical and physical and bio-geochemical aspects of the climate system as well as in the model’s response to external forcing. The phrase ‘model uncertainty’ is a common term in the climate change literature, but different studies use the phrase in different senses: some use it to represent the range of behaviours observed in ensembles of climate model (model spread), while others use it in more comprehensive senses (see Sections 9.2, 11.2 and 12.2). Model spread is often used as a measure of climate response uncertainty, but such a measure is crude as it takes no account of factors such as model quality (Chapter 9) or model independence (e.g., Masson and Knutti, 2011Masson, D., and R. Knutti, 2011: Climate model genealogy. Geophys. Res. Lett., 38, L08703.; Pennell and Reichler, 2011Pennell, C., and T. Reichler, 2011: On the effective number of climate models. J. Clim., 24, 2358–2367.), and not all variables of interest are adequately simulated by global climate models. To maintain a degree of terminological clarity this report distinguishes between ‘model spread’ for this narrower representation of climate model responses and ‘model uncertainty’ which describes uncertainty about the extent to which any particular climate model provides an accurate representation of the real climate system. This uncertainty arises from approximations required in the development of models. Such approximations affect the representation of all aspects of the climate including the response to external forcings. Model uncertainty is sometimes decomposed further into parametric and structural uncertainty, comprising, respectively, uncertainty in the values of model parameters and uncertainty in the underlying model structure (see Section 12.2). Some scientific research areas, such as detection and attribution and observationally-constrained model projections of future climate, incorporate significant elements of both observational and model-based science, and in these instances both sets of relevant uncertainties need to be incorporated. Scenario uncertainty refers to the uncertainties that arise due to limitations in our understanding of future emissions, concentration or forcing trajectories. Scenarios help in the assessment of future developments in complex systems that are either inherently unpredictable, or that have high scientific uncertainties (IPCC, 2000). The societal choices defining future climate drivers are surrounded by considerable uncertainty, and these are explored by examining the climate response to a wide range of possible futures. In past reports, emissions scenarios from the SRES (IPCC, 2000) were used as the main way of exploring uncertainty in future anthropogenic climate drivers. Recent research has made use of Representative Concentration Pathways (RCP) (van Vuuren et al., 2011avan Vuuren, D. P., et al., 2011a: RCP2.6: Exploring the possibility to keep global mean temperature increase below 2°C. Clim. Change, 109, 95–116., 2011bvan Vuuren, D. P., et al., 2011b: The representative concentration pathways: An overview. Clim. Change, 109, 5–31.). Internal or natural variability, the natural fluctuations in climate, occur in the absence of any RF of the Earth’s climate (Hawkins and Sutton, 2009Hawkins, E., and R. Sutton, 2011: The potential to narrow uncertainty in projections of regional precipitation change. Clim. Dyn., 37, 407–418.). Climate varies naturally on nearly all time and space scales, and quantifying precisely the nature of this variability is challenging, and is characterized by considerable uncertainty. The analysis of internal and forced contributions to recent climate is discussed in Chapter 10. The fractional contribution of internal variability compared with other forms of uncertainty varies in time and in space, but usually diminishes with time as other sources of uncertainty become more significant (Hawkins and Sutton, 2009; see also Chapter 11 and FAQ 1.1). In the WGI contribution to the AR5, uncertainty is quantified using 90% uncertainty intervals unless otherwise stated. The 90% uncertainty interval, reported in square brackets, is expected to have a 90% likelihood of covering the value that is being estimated. The value that is being estimated has a 5% likelihood of exceeding the upper endpoint of the uncertainty interval, and the value has a 5% likelihood of being less than that the lower endpoint of the uncertainty interval. A best estimate of that value is also given where available. Uncertainty intervals are not necessarily symmetric about the corresponding best estimate. In a subject as complex and diverse as climate change, the information available as well as the way it is expressed, and often the interpretation of that material, varies considerably with the scientific context. In some cases, two studies examining similar material may take different approaches even to the quantification of uncertainty. The interpretation of similar numerical ranges for similar variables can differ from study to study. Readers are advised to pay close attention to the caveats and conditions that surround the results presented in peer-reviewed studies, as well as those presented in this assessment. To help readers in this complex and subtle task, the IPCC draws on specific, calibrated language scales to express uncertainty (Mastrandrea et al., 2010Mastrandrea, M. D., et al., 2010: Guidance notes for lead authors of the IPCC Fifth Assessment Report on Consistent Treatment of Uncertainties. Available at http://www.ipcc.ch (accessed 07-10-2013).), as well as specific procedures for the expression of uncertainty (see Table 1.2). The aim of these structures is to provide tools through which chapter teams might consistently express uncertainty in key results. Notes Navigation ES 1.1 1.2.1 1.2.2 1.2.3 1.3 1.3.1 1.3.2 1.3.3 1.3.4 1.3.4.1 1.3.4.2 1.3.4.3 1.4.1 1.4.2 1.4.3 1.4.4 1.5 1.5.1 1.5.2 1.6 Box 1 FAQ Refs